IJMGE, VOL.50, NO.

2, (2016) 211–218

International Journal of Mining and Geo-Engineering

Blast fragmentation analysis using image processing

F. Sereshki a*, S. M. Hoseini a, M. Ataei a
a
Faculty of mining, petroleum and geophysics engineering, University of Shahrood, Shahrood, Iran

ARTICLE HISTORY
Received 24 Mar 2016, Received in revised form 13 Jul 2016, Accepted 13 Jul 2016

ABSTRACT

In this research, first of all, the common problems in fragmentation analysis are reviewed with regard to the reliability and rapidity of the
evaluation. Then, the available methods used for evaluation of blast results are described. The usual errors especially in recognizing rock
fragments in computer-aided methods, and also, the importance of accurate determination of fragment sizes in image analysis methods are
illustrated. After reviewing previous research work in this area, an algorithm is proposed for automated determination of rock particles’
boundary in Matlab software. This method can automatically determine the particles boundary in very short time. The results of proposed
method are compared with those of Split Desktop and GoldSize software packages in two automated and manual states. Comparing the curves
extracted from different methods reveals that the proposed approach is accurately applicable in measuring the size distribution of laboratory
samples, while the manual determination of boundaries in conventional software packages is very laborious and time-consuming; and
moreover the results of automated netting of fragments are very different from real values due to the error in separation of the objects.
Keywords: Auto delineation; Fragmentation measurement; GoldSize; Matlab software; SplitDesktop

for production blasting, this method is costly, time-consuming and

1. Introduction
The primary objective of rock blasting is to attain a successful
fragmentation. The classification and size distribution of muckpile are
the critical components of managing any blasting operation. The
fragmentation affects all downstream operations including loading,
hauling and crushing; and can be used to minimize these costs [1-8]. In
order to effectively control and optimize the process it is essential that
a rapid and reliable technique for assessing the degree of fragmentation
to be adopted. This is also important from the design point of view
where various different types of explosives and blast designs can be
quickly and efficiently analyzed [1]. Reliable evaluation of
fragmentation is a critical mining issue and quick and accurate
measurements of size distribution are essential to manage fragmented
rock. It can be used to optimize all blasting parameters to reduce costs
[9].
Although many techniques have been used for evaluation of the
fragmentation, but their effectiveness in actual operations has seldom
been documented [1]. There are several methods of size distribution
measurement which fall under two broad categories; direct and indirect
methods [5]. Sieving (or screening) is a direct and accurate method of
evaluation of size distribution of particles and fragmentation; However,
inconvenient [4]. Hence, indirect methods, which are observational,
empirical or digital methods have been developed. Observational
methods include the visual observation of muckpile immediately
following the blasting. It is widely used by blasting engineers to arrive at
an approximation. In some empirical models such as Larsson’s equation,
SveDeFo formula, KUZ-RAM model, etc, blasting parameters are
considered to determine the size distribution of blasted rock [5, 10, 7,
11].
In this regard, another technique is using image processing programs
which have been developed and have made rapid and accurate blast
fragmentation distribution assessment possible [5, 12]. Some of these
systems include IPACS, TUCIPS, FRAGSCAN, CIAS, GoldSize,
WipFrag, SPLIT, PowerSieve and Fragalyst[4, 12]. Split Desktop,
WipFrag, FragScan and GoldSize are the most popular software
packages based on 2D image processing for performing size distribution
analysis of the blasted rock blocks [12].
In addition to the aforementioned software packages which are
employed in mining industry, the use of image processing and machine
vision techniques is also on the rise. These techniques are utilized for
detection of rock or clod particles, mostly in mineral processing plants
(for detecting particles on the moving belt conveyor) or metallurgical
processes, or even in remote sensing. However, these digital methods
have inherent limitations, which adversely affect accuracy, precision,
and reproducibility of measurement results. This stems from the fact

* Corresponding author. Tel.: +98-2332395509, Fax: +98-2332395509.

E-mail address: farhang@shahroodut.ac.ir (F. Sereshki).

DOI: 10.22059/ijmge.2016.59831
Journal Homepage: ijmge.ut.ac.ir
212 F. Sereshki et al./ IJMGE 50-2 (2016) 211–218
that there are myriads of variables, which influence the outcome of the
measurements [14, 15]. Errors start with the imaging process and even
more errors are introduced in the digital processing stage, where blocks
may be miss-identified [14].
The image analysis techniques rely on first obtaining a digitized
outline of each individual particle from a photograph by manual
digitizing or automatic netting by CPU, then measuring the size and
shape parameters of the particles by computer [16]. Manual digitizing is
very time-consuming but it allows human interpretation of indistinct
particle edges (that would otherwise cause disintegration), so
measurement errors are reduced [1]. The time-consuming nature of
manual digitizing will prevent from fast evaluation as an advantage for
the computer software.
On the other hand, to ensure the validity of results in automatic
netting of particles’ boundary, there should be an appropriate contrast
between the particles. Therefore, image analysis methods work best
under controlled conditions like over moving conveyor belts, where
camera angles and distances can be held constant and lighting can be
controlled [2, 3]. However, in practice, it is not possible to prepare the
above conditions in big mines.
Because most imaging processing algorithms input data according to
the shadows between fragments, using these methods to delineate
individual blocks, as well as highly textured or multicolor fragments is
inclined to confuse the block delineation algorithms, resulting in falsely
identified fragment edges, and missed real fragment edges. In terms of
color characteristics, the lighter the color of the rock, the easier it is for
the edge delineation algorithms to correctly identify the edges. Even
though this is true, it is still possible to image materials of all colors from
white quartzite to black coal. Equally problematic areas are assemblages
where individual fragments exhibit mottling, or color density variations.
These also typically result in poor fragment delineation. Surface texture
on the fragments also tends to confuse the edge delineation algorithms,
in the same way that color differences do. In addition, fragments with
void spaces are often difficult to deal with. The situation is worse in the
case of washed and wet fragments, as this highlights color differences
[3, 17]. Therefore, analyzed particle size can be over-divided or
combined, which means larger particles can be divided into smaller
particles and smaller particles can be grouped into larger particles. This
is a common problem in all image processing programs [4, 5, 12, 13, 18,
and 19].
On one hand the manual digitizing will increase the approximation
accuracy and at the same time the required time; and on the other hand,
although the automated determination of particles’ boundary is done in
a matter of seconds, the approximation results usually don't have high
accuracy, and in most cases the results are not reliable. So in software
automatic determination of boundaries, there is often a need for a
manual-edition stage which is time-consuming as well. By investigating
the aforesaid problem, Sudhakar et al. declared that in automatic
netting, the estimated average dimensions of the fragments are either
bigger or smaller than their real values [4].
Regarding the abovementioned problems in detecting rock particle
dimensions using the existing software, in this research, an appropriate
algorithm is developed by using the image processing methods and
presented for detection of rock particles; then the obtained results are
compared with those of SplitDesktop and GoldSize. At first, by
introducing image separating techniques in image analyzing methods,
the previous studies in this filed are reviewed, and then, the techniques
employed in the current research and its results are illustrated.
Afterwards, the way of working with SplitDesktop and GoldSize
software is briefly explained, and their output results are demonstrated.
2. Literature review
The image analysis systems make use of segmentation techniques for
automated detection of particles’ boundary. This technique divides the
image into homogeneous areas. Edge detection and Morphology-based
methods are two examples of this technique [19]. Rock particle image
segmentation is typically the first and most difficult step. All subsequent
interpretation tasks, including particle size, shape and texture analysis,
rely heavily on the quality of the segmentation results [20]. The level of
success in segmentation algorithms will determine the accuracy of the
estimated size distribution [15]. Since rock particle images vary from
one to another, it is difficult or even impossible to design and develop
one segmentation algorithm for all kinds of rock particle images. The
presented segmentation algorithms were developed for just several
types of rock particle images with certain characteristics with respect to
segmentation [20].
In 2006 Al-Thyabat and Miles used images of separated rock particles
to evaluate the efficiency of measuring two different dimensions of the
particles with the aim of attaining particle size distribution. Also, in
order to separate touching particles in images, they utilized the
Watershed algorithm [21].
In 2011 Thurley employed a morphological edge detection strategy to
draw the boundary of limestone particles on a conveyor belt. In this
work, 3-D data were used [17]. In 2012, Zelin et al. applied a series of
pre-analyzing steps on the images using software to solve the
overlapping problem for coal particles on conveyor belt; in the first step,
they enhanced the primary image using Otsu method, afterwards the
grayscale image was created. Also, exponential high pass filter and
Fourier transform have been used in order to improve the images; and
the edges of the image were detected by morphological edge detection
[15]. In the same year Jemwa and Aldrich [19] investigated the problem
of predicting the size class of coal particles on a conveyor belt with
respect to the amount of fine size fraction, and for the purpose of
particle identification, they proposed a machine vision approach based
on textural characterization of images from coal material.
In 2014, Chimi et al. used a method based on a well-known
segmentation algorithm which is called watershed segmentation
method in order to detect clods in the soil; that is also applicable for
remote sensing [22]. In the same year Al-Thyabt et al. focused on the
difficulties raised in analyzing the images of coal particles on a moving
conveyor belt, such as, camera location, overlapping of particles, image
blurring, conveyor belt speed, dust generation and the treatment. After
image analysis and using Gaussian filter for the image enhancement,
they determined the coal particles’ boundary manually [23].
In 2013 and 2014 some other efforts were made in application of
image processing for estimating the particle size distribution. In these
research works, the particles’ boundaries were segmented manually, or
the image processing aspects were not taken into consideration [24, 25].
Although segmentation methods have been successfully utilized in
many object recognition systems, their use in particle size distribution
estimation of aggregate material is still a challenge due to the
abovementioned problems; and thus no general solution is available on
the market [17, 19, and 23].
3. Sample Preparation and Photography
In order to evaluate the capability of determining the fragmentation
size distribution with the approach developed using Matlab software,
and to compare the results with SplitDesktop and GoldSize results, the
laboratory samples of crushed rock were prepared by screening. The
preparation was based on Rosin-Rammler distribution function
(relation (1)) that is a function indicating the passing percentage of
material mass. At first, specified values were taken for n and d50, and the
weight percentage of each fraction was calculated according to Eq. (1)
to acquire a 10kg sample. In Figure 1 one can see the image of first
sample in two distributed and pile states with d50 equal to 38mm and n
equal to 2. Table 1 shows the particle size and weight percentage of
fractions for sample 3.
𝑥 𝑛
)
𝑅(𝑥) = 1 − 𝑒 −(𝑥𝑐 (1)
In (1), R(x) is the ratio of the mass passed through a screen having a
dimension of x; while xc is the scale parameter (size at which the fraction
passing is 63.9%), and n is the uniformity parameter [26].
 F. Sereshki et al./ IJMGE 50-2 (2016) 211–218 213

(a)
important to us. The edge pixels are those in which the image brightness
(b)
intensity varies, and the edges are a set of connected edge pixels related
to each other. The line can be considered as a part of edge in which the
background intensity on either sides of line is significantly higher or
lower than the line pixels’ intensity.
The edge detection is performed on basis of intensity changes. The
model under evaluation in this research is edge detection based on the
differences between two levels of intensity. Discovery of intensity
changes for the edge detection purpose can be implemented using the
first and second derivatives. Equations (2) and (3) give the first and
second derivatives of the image.
f
Figure 1. Laboratory sample of crushed rock (n=1.4, d 50=19), (a): Pile  f ( x)  f ( x  1)  f ( x)
(b): Distributed. x (2)
2
Table 1: Size and weight of each fraction for the sample 3 (n=1.4, and d 50=19).  f
 f ( x  1)  f ( x)  f ( x  2)  2 f ( x  1)  f ( x)
Screen size Fraction Weight ratio of each Weight of each fraction 2

(mm) (mm) fraction (in grams) out of 10kg  x (3)

0.15 0 - 0.15 0.000789 7.9 In order to calculate the edge power and direction in (x, y) position
0.3 0.15 - 0.3 0.001291 12.9 of the image f, the gradient operator ( f ) in its vector form is used as
0.6 0.3 - 0.6 0.003400 34.0
1.2 0.6 - 1.2 0.008917 89.2 Eq. (4). The amplitude and direction of f are also calculated using Eq.
2.36 1.2 - 2.36 0.022293 222.9 (5) and (6).
4.75 2.36 - 4.75 0.058044 580.4
9.51
12.7
4.75 - 9.51
9.51 - 12.7
0.136555
0.094601
1365.5
946.0
 
 f 
f  grad ( f )  gg   fx  x
(4)
 y 
y

19 12.7 - 19 0.174110 1741.1

25.4 19 - 25.4 0.146803 1468.0
38 25.4 - 38 0.192659 1926.6
51 38 - 51 0.097350 973.5
63.5 51 - 63.5 0.039755 397.6
125 > 63.5 0.023432 234.3

4. Proposed Method for Automated Detection of Rock

Particles

The procedure of the proposed approach is illustrated in Figure 2. As

shown, initially the input images are transmitted to the grayscale space.
The grayscale image is a monochrome image which indicates the
brightness degree of the image.
According to different lighting conditions and camera location in
imaging of each sample, every image will require its own specific image
processing, so that different results were obtained for images by
applying filters with definite dimensions and thresholds. Thus, to reduce
this effect, the form shown in Figure 3 was considered for the data input
in order to be able to evaluate the effect of filter dimension and
threshold on various images with the purpose of acquiring optimal
dimension and threshold values.
After generating the grayscale image, the median filter with the user-
defined dimensions is applied to the image. This filter is a low-pass filter
that due to its nature requires more processing time compared to other
filters. Its working principal is in the way that it sorts the whole
neighborhoods in an ascending order, then, selects the middle element
of the sorted values and substitutes in the central pixel. Subsequently, in
order to smooth the image, the Gaussian filter having 3×3 dimensions is
utilized.
Afterwards, various segmentations were applied, and the Sobel filter
is used. The Sobel masks outperform in noise removal. The
segmentation divides the image into its constituent parts or objects. The Figure 2. Procedure of the proposed method for detecting size of rock particles.
degree of division depends on the problem at hand, in other words the 2 2

segmentation is terminated when the desirable objects or regions are M ( x, y)  mag(f )  g  g

identified. Most of the segmentation algorithms are based on one of the
x y
(5)
two main properties of intensity, i.e. discontinuity and similarity. In the g 
first category, the methodology is to divide the image based on sharp  ( x, y )  tan 1  x 
 g y  (6)
changes of intensity like edges. The basis of main method in the second
category is based upon the division of image into regions which are Achieving the gradient of image requires calculation of partial
similar according to a set of predefined criteria. f f
Line and edge detection is a kind of segmentation which is based on derivatives of x and y in the position of each pixel in the image.
detecting sharp local changes in the intensity. The points, lines, and Since the relevant quantities are digital ones, digital approximation of
edges are the main components of image which their separation is partial derivatives on the output of a neighborhood around a point is
214 F. Sereshki et al./ IJMGE 50-2 (2016) 211–218

g g method acts based on the shape of objects existing within the image. The
required. Relations (7) and (8) represent the discrete form of x and y morphology consists of some basic primary operators, such as Dilation,
that can be implemented using image filtering through one-dimensional Erosion, Opening, and Closing; and the other morphological operators
masks. are defined using the primary ones. In this work, for enhancement the
f ( x, y ) quality of the images, the “Opening” operator is employed after the
g   f ( x  1, y )  f ( x, y )
x x (7) application of filters. For the binary and grayscale images, these
f ( x, y ) operators are defined in two forms. In the current paper, the
g   f ( x, y  1)  f ( x, y ) abovementioned operators are used for the grayscale images.
y y (8) By identifying the connected regions, and labeling of each identified
The edge detection technique of Canny is more efficient than the pixel, the rock particles are detected, and the area of each particle is
Sobel method. In other words, the Canny algorithm has lower error rate. determined according to the number of pixels within the scale balls, and
But the improvements of the Canny technique comes at the price of then, the result is stored in an “Excel” file.
more complex implementation compared to Sobel method and longer The images resulted from the proposed image processing approach
execution time. have been illustrated in two pile and distributed states in Figures 4 and
After detecting lines, another median filter with dimensions of 5×5 is 5.
applied, and the morphological operators are used. In image processing,
the morphology is used as a tool to extract the image components. This

Figure 3. Form of determining the threshold and dimensions of median filter.

Figure 4. Output images of the proposed approach for distributed laboratory sample, (a): grayscale image after detection of lines (b): threshold image (c): applying the
second median filter (d): employing morphological operators.

Figure 5. Output images of the proposed approach for pile laboratory sample, (a): grayscale image after detection of lines (b): threshold image (c): application of the
second median filter (d): employing morphological operators.
 F. Sereshki et al./ IJMGE 50-2 (2016) 211–218 215

5. SplitDesktop
SplitDesktop, which is based on 2-D images processing, is one of the
most popular software packages for blast fragmentation analysis. This
software uses images captured with two objects (balls) that are used as
scaling tools. After the scaling process, the software delineates images
automatically with an image filter as depicted in Figure 6. As it can be
seen, the boundaries have not been drawn correctly, and the results
(particle size distribution curve) are completely different from reality.
Hence, regarding the possibility of manual edition of software automatic
delineation, the particles boundary was drawn manually from the
beginning that took a long time (Figure 7). After determining the
particles in the previous stage, the software draws the particle size
distribution curve as the output.
(a) (b)
boundaries are drawn by the user from the beginning. Figure 8 shows
the working environment of this software in which the boundaries have
been segmented.
7. Comparison of Results
The figures 9, 10, 11 illustrate the particle size distribution curve for
the samples 1, 2, and 3, in two distributed and pile states. In each curve,
the particle size distribution is resulted from the screening analysis,
GoldSize and Split software, and the proposed approach in Matlab
software. The obtained curve from Split is the result of manual
delineation in SplitDesktop, and the Split-auto refers to delineation by
the software automatically.

Figure 6. Automated delineation of particle boundaries in SplitDesktop software,

(a): Pile (b): Distributed.

(a) (b)

Figure 7. Manual delineation of particle boundaries in SplitDesktop software,

(a): Pile (b): Distributed. Figure 8. Working environment of GoldSize software, right: delineation state,
left: scaling state.

6. GoldSize

In this software, the particle sizing of fragmented rock is also

evaluated based on 2-D digital images. In this system, the particle

Figure 9. Size distribution curves for sample 1 by different methods (n=2, d50=38mm).
216 F. Sereshki et al./ IJMGE 50-2 (2016) 211–218

By using of experimental expression (9) which is suggested by
Sudhakar [4], the uniformity parameter (n) can be calculated and
compared with the real value based on d50 (size at which the passing
fraction is 50%) and d80 (size at which the passing fraction is 80%).
𝑛 = 0.842/(𝐿𝑛 𝑑80 − 𝐿𝑛 𝑑50 )
In Table 2, the values of d50 and d80, (recorded from the curve) and
also, n (calculated by Eq. (9)) for the different methods applied are
presented.
(9)

Figure 10. Size distribution curves for sample 2 by different methods (n=1.1, d50=25.4mm).

Figure 11. Size distribution curves for sample 3 by different methods (n=1.4, d50=19mm).

By evaluating the curves and the table information, one can deduct pronounced as the particles decrease in size (from sample 1 to 3). By
the following results: investigating the delineated images, it was observed that in some cases,
Except in one case, the results obtained from the automated a set of fine rock particles are regarded as a single rock particle and also,
delineation in SplitDesktop software do not conform to the real size the boundaries of large particles have not been determined correctly.
distribution (screening results). This incompatibility becomes more
Table 2. Comparison of particle size distribution results for the three samples.
Sample 𝒅𝟓𝟎 𝒅𝟖𝟎 n
no. Sc GS S M Sa Sc GS S M Sa Sc GS S M Sa
1 D 38 51 38.1 39 50.5 67 84.7 65.3 59.1 71.3 2 1.7 1.6 2 2.4
P 38 41 35.8 40 39.4 67 56 55.4 61.9 53.5 2 2.7 1.9 1.9 2.7
2 D 25.4 26.5 24.1 34 53.1 54 51 53.2 59 76.1 1.1 1.3 1.1 1.5 2.3
P 25.4 31 39.3 30 42.9 54 71 84.4 64 69 1.1 1 1.1 1.1 1.8
3 D 19 28 20 27 94.7 35 53 41.6 56 131.3 1.4 1.3 1.1 1.2 2.6
P 19 27 23.8 27 44.7 35 60 46.4 61 67.5 1.4 1.1 1.3 1 2
D: Distributed, P: Pile, Sc: screening analysis, GS: GoldSize, M: Matlab, Sa: Split-auto
 F. Sereshki et al./ IJMGE 50-2 (2016) 211–218 217

 Although in the “Split-auto” case the boundary of rocks is SplitDesktop is easier to work with than Goldsize, and also the
detected by the software, for separating the background image correction of user-induced errors is better possible within the
and drawing the boundaries of surrounding rocks, a manual former one.
edition stage is required.  The results of the proposed method are acceptable regarding
 The results of manual determination of boundaries in the automated determination of the particles boundaries; and
SplitDesktop and GoldSize software packages are better in also, with regard to the results of manual determination in both
distributed state than in the pile one, which is due to of the two software packages, rock particles are not detected
observation of all the rock fragments and determination of the correctly when the particles are fine.
boundaries of each one.
 In the finer sample (sample 3), despite the manual delineation,
8. Conclusion
both SplitDesktop and GoldSize have estimated larger values
for the particle size distribution. In these software packages, it
In this paper, in order to simplify and accelerate the fragmentation
is not possible to determine all the fine particles due to the
measurement of the blasted rock, the Matlab software was employed. To
limitation on the image contrast. In GoldSize software, the fine
validate the results, the fragmentation of laboratory samples with a
particles are not determined correctly, and in SplitDesktop,
specified size distribution, was determined. For further validation, the
only the boundaries of the fine particles are detected and
results obtained from Matlab were compared with those of SplitDesktop
measured.
and GoldSize software packages (errors of these methods are shown in
 It is worth mentioning that in the manual delineation step, table 3).
Table 3. Errors in proposed method, SplitDesktop and GoldSize for calculation of 𝐝𝟓𝟎 , 𝐝𝟖𝟎 and n (%).
Sample 𝒅𝟓𝟎 𝒅𝟖𝟎 n
no. Sc GS S M Sa Sc GS S M Sa Sc GS S M Sa
1 D -20.0 0.0 20.0 34.2 0.3 2.6 32.9 26.4 -2.5 -11.8 6.4 -15.0 -20.0 0.0 20.0
P -5.0 -5.0 35.0 7.9 -5.8 5.3 3.7 -16.4 -17.3 -7.6 -20.1 35.0 -5.0 -5.0 35.0
2 D 0.0 36.4 109.1 4.3 -5.1 33.9 109.1 -5.6 -1.5 9.3 40.9 18.2 0.0 36.4 109.1
P 0.0 0.0 63.6 22.0 54.7 18.1 68.9 31.5 56.3 18.5 27.8 -9.1 0.0 0.0 63.6
3 D -21.4 -14.3 85.7 47.4 5.3 42.1 398.4 51.4 18.9 60.0 275.1 -7.1 -21.4 -14.3 85.7
P -7.1 -28.6 42.9 42.1 25.3 42.1 135.3 71.4 32.6 74.3 92.9 -21.4 -7.1 -28.6 42.9

 The conclusions that can be drawn from this research are as
follows:
 Although SplitDesktop is one of the most widely-used software
packages for the blast-induced fragmentation analysis, there is
a need for its manual correction, and hence, it is very laborious
and time-consuming due to incorrect delineation of fragmented
rocks and error in the rock detection in automatic mode.
 In GoldSize software, the rocks boundaries can only be
segmented manually and by taking a lot of time.
 Determining the boundary of the particles is the main and most
important step of image processing that the subsequent
calculations are dependent upon.
 In this research, the edge detection is performed using the Sobel
method by converting the colored images to grayscale ones and
by applying the median filter; Moreover, thresholding, re-
applying the median filter, and morphology operators are
employed to acquire the desired image. In addition, by
designing a data input form, it was possible to determine the
filter dimensions and also proper threshold for each image, with
regard to the difference of required processes for each image
which is due to the variation of environmental conditions.
 Comparing the results of different methods and errors of these
methods demonstrated that the proposed approach, is
accurately applicable in measuring the size distribution of
laboratory samples, especially thanks to its ability to
automatically determine particle boundaries in the shortest
time.
REFERENCES
[1] Hunter, G.C., McDermott, C., Miles, N.J., Singh, A. & Scoble, M.J.
(1990). A review of image analysis techniques for measuring
blast fragmentation. Mining Science and Technology, 11(1), 19-
36.
[2] Palangio, T.W., Palangio, T.C. & Maerz, N. (2005). Advanced
automatic optical blast fragmentation sizing and tracking.
European Federation of Explosives Engineers Conference
Proceedings, Brighton, 259-267.
[3] Maerz, N.H. & Palangio, T.W. (2004). Post-muckpile, pre-primary
crusher, automated optical blast fragmentation sizing. Fragblast,
8(2), 119–136.
[4] Sudhakar, J., Adhikari, G.R. & Gupta, R.N. (2006). Comparison of
fragmentation measurements by photographic and image
analysis techniques. Rock Mechanics and Rock Engineering, 39
(2), 159-168.
[5] Siddiqui, F.I., Ali Shah, S.M. & Behan, M.Y. (2009). Measurement
of size distribution of blasted rock using digital image
processing. Engineering Science, 20(2), 81-93.
[6] Singh, A., Scoble, M., Lizotte, Y. & Crowther, G. (1991).
Characterization of underground rock fragmentation.
Geotechnical and Geological Engineering, 9(2), 93-107.
[7] Chakraborty, A.K., Raina, A.K., Ramulu, M., Choudhury, P.B.,
Haldar, A., Sahu, P. & Bandopadhyay, C. (2004). Parametric
study to develop guidelines for blast fragmentation
improvement in jointed and massive formations. Engineering
Geology, 73(1-2), 105– 116.
[8] Mohammadnejad, M., Gholami, R., Sereshki, F. & Jamshidi, A.
(2013). A new methodology to predict backbreak in blasting
operation, International journal of rock mechanics and mining
sciences, 60, 75-81.
[9] Maerz, N.H., Palangio, T.C. & Franklin, J.A. (1996). WipFrag image
based granulometry system. Proceedings of the FRAGBLAST 5
Workshop on Measurement of Blast Fragmentation, Montreal,
Quebec, Canada, 91-99.
[10] Lopez Jimeno, Carlos, (1995). Drilling and blasting of rocks,
Rotterdam A.A. Balkema.
[11] Thornton, D., Kanchibolta, S. & Esterle, J. (2001). A fragmentation
model to estimate ROM size distribution of soft rock types.
Proceedings of the Twenty-Seventh Annual Conference On
Explosives and Blasting Technique, Cleveland, Ohio, 1, 41-53.
218 F. Sereshki et al./ IJMGE 50-2 (2016) 211–218
[12] Kim, K. (2006). Blasting design using fracture toughness and
image analysis of the bench face and muckpile. M.S, thesis,
Virginia polytechnic institute and state university, Blacksburg.
[13] Han, J-H. & Song, J-J. (2014). Statistical estimation of blast
fragmentation by applying stereophotogrammetry to block piles.
International Journal of Rock Mechanics & Mining Sciences, 68,
150–158.
[14] Maerz, N.H. & Zhou, W. (1998). Optical digital fragmentation
measuring systems – inherent sources of error, The
International Journal for Blasting and Fragmentation, 2(4), 415-
431.
[15] Zelin, Z., Jianguo, Y., Lihua, D. & Yuemin, Z. (2012). Estimation
of coal particle size distribution by image segmentation.
International Journal of Mining Science and Technology, 22,
739–744.
[16] Wang, W. (2006). Size and shape measure of particles by image
analysis. 11th International Workshop, IWCIA Proceedings,
Berlin, Germany, 4040, 253 – 262.
[17] Thurley, M.J. (2011). Automated online measurement of
limestone particle size distributions using 3D range data. Journal
of Process Control, 21(2), 254–262.
[18] Maerz, N.H. & Zhou, W. (1999). Calibration of optical digital
fragmentation measuring systems. FRAGBLAST 6, Sixth
International Symposium For Rock Fragmentation By Blasting,
Johannesburg, South Africa, 125-130.
[19] Jemwa, G.T. & Aldrich, C. (2012). Estimating size fraction
categories of coal particles on conveyor belts using image texture
modeling methods. Expert Systems with Applications, 39(9),
7947–7960.
[20] Weixing, W. (2008). Rock Particle Image Segmentation and
Systems, Pattern Recognition Techniques, Technology and
Applications, Peng-Yeng Yin (Ed.), ISBN: 978-953-7619-24-4,
InTech, Available from:
http://www.intechopen.com/books/pattern_
recognition_techniques_technology_and_applications/rock_par
ticle _image_segmentation _ and _ systems.
[21] Al-Thyabat, S. & Miles, N.J. (2006). An improved estimation of
size distribution from particle profile measurements. Powder
Technology, 166(3), 152–160.
[22] Chimi, C.O., Hegarat, M.S.L., Vannier, E., Taconet, O. &
Dusseaux, R. (2014). Automatic clod detection and boundary
estimation from Digital Elevation Model images using different
approaches. Catena, 118, 73–83.
[23] Al-Thyabat, S., Miles, N.J. & Koh, T.S. (2007). Estimation of the
size distribution of particles moving on a conveyor belt. Minerals
Engineering, 20(1), 72–83.
[24] Hamzeloo, E., Massinaei, M. & Mehrshad, N., (2014). Estimation
of particle size distribution on an industrial conveyor belt using
image analysis and neural networks. Powder Technology, 261,
185–190.
[25] Zhang, Z., Yang, J., Su, X. & Ding, L. (2013). Analysis of large
particle sizes using a machine vision system, Physicochemical
Problems of Mineral Processing, 49(2), 397−405.
[26] Sanchidrian, J.A., Ouchterlony, F., Moser, P., Segarra, P. & Lopez,
L.M. (2012). Performance of some distributions to describe rock
fragmentation data. International Journal of Rock Mechanics &
Mining Sciences, 53, 18–31.